I have my notebook mounted on my treadmill. Every night after my toddler goes to bed I game while walking for an hour or two. It's set for a 5 degree incline and 1.5 MPH. It doesn't affect my playing abilities and I burn a lot of calories. Also, my heart rate doesn't get so high as to affect getting to sleep afterward. As a bonus, if I want to play more (for those long WoW raids) I burn more calories!
In some of the reading I've done on game AI I was surprised to find how little was 'traditional/academic' like neural networks and fuzzy logic and such. It appeared to be more mundane like pathfinding, facial/body expressions, reaction adjustment, tactics, etc. I could easily see a 'pathfinding' engine that could set the paths for millions of 'creatures' or a very deep facial/body expression engine that might take advantage of state history or current game circumstances also for millions of NPCs or perhaps an ecosystem machine that was better than a series of waypoints for each creature that could keep track of entire game levels. There are perhaps accelerations that can be done for game physics as well. That was part of what I meant by push/pull. The 3D guys developed algorithms for lighting, ray tracing, fog, etc which the APIs implemented. These were not originally envisioned by simple triangle drawing and texture mapping.
That would certainly help significantly but wasn't true for the first 3D cards. Each of the chip makers provided their own non-standard APIs and drivers which game developers would test with and have special hardware specific switches for. I remember turning in my Rendition card for a 3dfx card because there was specific 3dfx port for the game Descent. Of course things accelerated more swiftly when the market thinned down and settled on DirectX and OpenGL. A push/pull relationship seemed to develop between HW features and API features which (IMHO) was an overall benefit to the entire 3D/gaming market and caused the best features from the differnent vendors to be implemented by everyone. I would hope something like would happen to any HW accelerated AI chip. For example new pathfinding algorithms might develop faster than NPC reaction algorithms. Or HW would open the door for new features which simply can't be done under SW/real time constraints.
In one of my past lives I asked an architect of one of an early generation 3D graphics chip if there might be a market for AI accelerator chips that could provide a similar leap for AI that 3D accelerators added to games. I noted the increased amount of CPU time spent by games on AI as the amount of time spent on graphics decreased. He brushed aside my comment so I'd like to ask it again here.
I'd envision the device would not be 'intelligent' but rather accelerate, keep state and perhaps otherwise add depth, randomness and the ability to have 1000's more distinct AIs than normal. Algorithms would be accessable through AI APIs (like OpenGL or DirectX does for 3D) that developers could use rather than re-invent the AI wheel for each game.
I admit I haven't read any AI books to completion so the comment is directed at AI algorithm developers and implementers who might wish there was some way to do a lot more things a lot faster without having to prune CPU hungry AI code.
Slashdot Mirror
I have my notebook mounted on my treadmill. Every night after my toddler goes to bed I game while walking for an hour or two. It's set for a 5 degree incline and 1.5 MPH. It doesn't affect my playing abilities and I burn a lot of calories. Also, my heart rate doesn't get so high as to affect getting to sleep afterward. As a bonus, if I want to play more (for those long WoW raids) I burn more calories!
In some of the reading I've done on game AI I was surprised to find how little was 'traditional/academic' like neural networks and fuzzy logic and such. It appeared to be more mundane like pathfinding, facial/body expressions, reaction adjustment, tactics, etc. I could easily see a 'pathfinding' engine that could set the paths for millions of 'creatures' or a very deep facial/body expression engine that might take advantage of state history or current game circumstances also for millions of NPCs or perhaps an ecosystem machine that was better than a series of waypoints for each creature that could keep track of entire game levels. There are perhaps accelerations that can be done for game physics as well. That was part of what I meant by push/pull. The 3D guys developed algorithms for lighting, ray tracing, fog, etc which the APIs implemented. These were not originally envisioned by simple triangle drawing and texture mapping.
That would certainly help significantly but wasn't true for the first 3D cards. Each of the chip makers provided their own non-standard APIs and drivers which game developers would test with and have special hardware specific switches for. I remember turning in my Rendition card for a 3dfx card because there was specific 3dfx port for the game Descent. Of course things accelerated more swiftly when the market thinned down and settled on DirectX and OpenGL. A push/pull relationship seemed to develop between HW features and API features which (IMHO) was an overall benefit to the entire 3D/gaming market and caused the best features from the differnent vendors to be implemented by everyone. I would hope something like would happen to any HW accelerated AI chip. For example new pathfinding algorithms might develop faster than NPC reaction algorithms. Or HW would open the door for new features which simply can't be done under SW/real time constraints.
In one of my past lives I asked an architect of one of an early generation 3D graphics chip if there might be a market for AI accelerator chips that could provide a similar leap for AI that 3D accelerators added to games. I noted the increased amount of CPU time spent by games on AI as the amount of time spent on graphics decreased. He brushed aside my comment so I'd like to ask it again here. I'd envision the device would not be 'intelligent' but rather accelerate, keep state and perhaps otherwise add depth, randomness and the ability to have 1000's more distinct AIs than normal. Algorithms would be accessable through AI APIs (like OpenGL or DirectX does for 3D) that developers could use rather than re-invent the AI wheel for each game. I admit I haven't read any AI books to completion so the comment is directed at AI algorithm developers and implementers who might wish there was some way to do a lot more things a lot faster without having to prune CPU hungry AI code.